Closed WadeBarnes closed 8 months ago
I created a search-engine image off https://github.com/bcgov/aries-vcr/pull/759 and smoke tested in dev
and test
before deploying to prod
. Performing some load testing before leaving it over the weekend to collect some metrics.
Disabling the Solr caches had no effect. If anything the OOM restart issues became worse following the change, though this is just speculation and the increased number of issue could be attributed to query patterns and volume. Some metrics collected here; https://github.com/bcgov/orgbook-configurations/pull/137#issuecomment-1764701163.
This actually helped reduce resource usage. It was actually unbound queries (https://github.com/bcgov/aries-vcr/issues/762) that were the main issue.
As discussed some of the articles listed here caching is only useful when there are repeated queries for the same information, otherwise they can be detrimental to memory use. Queries in OrgBook tend to be random (Humans) or sequential (Spiders and Bots), so there are few, if any, cache hits. This could be, at least part of, the cause for the high memory usage we're seeing in
prod
.Try disabling the caches as recommended to see the impact on memory use.
This issue is related to the investigation happening here; https://github.com/bcgov/orgbook-configurations/pull/137